Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (2 results)

  • Thesis
  • Peer Reviewed
UC Berkeley Electronic Theses and Dissertations (2022)

This dissertation seeks to develop novel data-driven algorithms to automatically construct simulation-suitable meshes from volumetric medical image data to enable high-throughput, large-cohort analyses of patient cardiac function from medical image data. Image-based cardiac modeling derives geometries from patient medical image data, and can simulate a variety of aspects of cardiac function, including electrophysiology, hemodynamics, and tissue mechanics to explore improvements in cardiovascular diagnoses and treatments, and the biomechanical underpinnings of diseases. However, creating accurate models of the heart from patient image data requires significant time and human efforts and is the primary challenge of image-based modeling.

Deforming-domain computational fluid dynamics (CFD) simulations of the intracardiac hemodynamics, in particular, require both the geometry and the deformation of the heart from a sequence of image snapshots throughout the cardiac cycle. In the first part of the thesis, we present a two-stage approach to automatically generate CFD-ready left ventricle (LV) models from patient image data. This approach first uses deep-learning-based automatic segmentation and then uses geometry processing algorithms to robustly create CFD-suitable LV models from image data.

In the remainder of this dissertation, we developed end-to-end deep-learning algorithms to directly construct the surface meshes of the whole heart from volumetric medical image data without the need for a multistage approach. These algorithms leverage shape templates, shape deformation methods, and regularization functions during optimization to generate meshes that are suitable for computational simulations. Namely, we trained a graph convolutional neural network to deform a pre-defined mesh template to fit the whole heart in a target 3D image volume. Various mesh deformation methods were combined with deep learning to create whole heart surfaces, including direct displacements on meshes, free-form deformation by control lattices as well as control-handle-based deformation using bi-harmonic coordinates. We demonstrated the application of our method in constructing a dynamic whole heart mesh from time-series CT image data to simulate the cardiac flow driven by the cardiac motion. We also demonstrated the capability to switch template meshes to accommodate different modeling requirements.

The algorithms developed in this thesis were able to construct a 4D dynamic simulation-suitable mesh of the heart within a minute on a standard desktop computer whereas prior methods require hours of human efforts and multiple programs. The code was implemented in Python, fully open-sourced, and can be conveniently executed from the command line.

    Cover page: Automated Model Construction for Image-Based Cardiac Computational Simulations
    • Article
    • Peer Reviewed
    UC Berkeley Previously Published Works (2021)
    Automated construction of surface geometries of cardiac structures from volumetric medical images is important for a number of clinical applications. While deep-learning-based approaches have demonstrated promising reconstruction precision, these approaches have mostly focused on voxel-wise segmentation followed by surface reconstruction and post-processing techniques. However, such approaches suffer from a number of limitations including disconnected regions or incorrect surface topology due to erroneous segmentation and stair-case artifacts due to limited segmentation resolution. We propose a novel deep-learning-based approach that directly predicts whole heart surface meshes from volumetric CT and MR image data. Our approach leverages a graph convolutional neural network to predict deformation on mesh vertices from a pre-defined mesh template to reconstruct multiple anatomical structures in a 3D image volume. Our method demonstrated promising performance of generating whole heart reconstructions with as good or better accuracy than prior deep-learning-based methods on both CT and MR data. Furthermore, by deforming a template mesh, our method can generate whole heart geometries with better anatomical consistency and produce high-resolution geometries from lower resolution input image data. Our method was also able to produce temporally-consistent surface mesh predictions for heart motion from CT or MR cine sequences, and therefore can potentially be applied for efficiently constructing 4D whole heart dynamics. Our code and pre-trained networks are available at https://github.com/fkong7/MeshDeformNet.
      Cover page: A deep-learning approach for direct whole-heart mesh reconstruction
      Top